arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.28397cs.CRcs.AIcs.CL
收藏

MAS-Guard:从单个智能体到多智能体大语言模型系统的统一安全护栏

MAS-Guard: Unified Guardrails from Individual Agents to Multi-Agent LLM Systems

Oracle AI(甲骨文人工智能)

Mohammad Kachuee, Phu Mon Htut, Sheng Li, Nikolaos Pappas, Miguel Ballesteros, Yassine Benajiba, Sujith Ravi, Dan Roth

AI总结:

本研究针对多智能体系统防护碎片化问题,提出MAS-Guard统一框架,结合本地A-Guard层与系统级智能体章程、控制平面,可降低71.5%-86.3%的攻击成功率,同时保障良性任务效用。

摘要:

智能体人工智能将大语言模型(LLMs)从生成响应扩展到规划与执行复杂工作流,大幅提升了其实用性,同时也使应用面临不安全行为、信息泄露和对抗性操纵的风险。多智能体系统通过专业化与协作进一步拓展了这些能力,但分布式权限、异构角色以及不透明或可能被篡改的参与者,使得安全防护更具挑战性。现有防御措施针对重要的个体威胁,但在本地执行、系统级交互和执行机制方面的覆盖是碎片化的。我们提出MAS-Guard,这一统一框架可填补针对包含可检测与不透明智能体的单智能体及多智能体AI应用的防护空白。在可检测智能体中,所提出的本地智能体护栏(A-Guard)层采用任务特定专家评估智能体事件;在系统层面,开发者拥有的智能体章程与控制平面结合了对路由、智能体能力和消息格式的确定性结构检查,以及对角色和消息内容的语义评估。这些互补层通过一组共享干预措施(允许、反思、确认、阻止)实现执行。在涵盖不同单智能体与多智能体场景的6个基准上,使用来自不同提供商的3种智能体LLM进行评估,结果显示,与未防护的智能体相比,MAS-Guard将整体攻击成功率降低了71.5%-86.3%,同时在良性任务上满足了效用参考标准。

英文摘要:

Agentic AI extends large language models (LLMs) from generating responses to planning and executing complex workflows, substantially expanding their utility while exposing applications to unsafe actions, information leakage, and adversarial manipulation. Multi-agent systems further expand these capabilities through specialization and collaboration, but distributed authority, heterogeneous roles, and opaque or potentially compromised participants make safeguarding even more challenging. Existing defenses address important individual threats but provide fragmented coverage across local execution, system-level interactions, and enforcement mechanisms. We introduce MAS-Guard, a unified framework that closes this gap for safeguarding single- and multi-agent AI applications containing both instrumentable and opaque agents. Within instrumentable agents, the proposed local Agent Guardrail (A-Guard) layer applies task-specific experts to assess agentic events. At the system level, developer-owned Agent Charters and a Control Plane combine deterministic structural checks of routes, agent capabilities, and message formats with semantic assessments of roles and message content. These complementary layers enable enforcement through a shared set of interventions: allow, reflection, confirmation, and block. Evaluated across six benchmarks covering diverse single- and multi-agent scenarios with three agent LLMs from different providers, MAS-Guard reduces the overall attack success rate by 71.5-86.3% relative to unguarded agents, while meeting the utility reference criterion on benign tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑